Tencent's rubric-aligned search lifts Yuanbao end-to-end answer quality by 8.52%
_reachsumit · x · 2026-11-18
A Tencent paper (accepted at KDD 2027) introduces rubric-aligned search, deployed in Yuanbao Search, shifting the pipeline's goal from per-document relevance to utility for downstream LLM reasoning.
- For each query, a coarse-to-fine rubric defines weighted positive and negative criteria for the document set, unifying retrieval, ranking and document selection under one supervision surface
- A lightweight labeler distills rubric judgments into per-document utility labels for a contrastive dual-encoder retriever and a multi-task ranker; a rubric-conditioned evaluator provides set-level rewards for a dual-branch RL selector (GRPO + best rollout distillation)
- Deployed online, it improves end-to-end answer quality by 8.52% over the production baseline
More from Research
- Voyager 1 Reaches One Light-Day From Earth, Signals Now Take 24 Hours — snowclipsed · 2026-11-18
- TCS Hiring Statement Urges Quality Over Quantity as AI Fuels Paper Arms Race — fortnow · 2026-11-18
- New paper: a sociocultural approach to linguistic creativity — abenitezburraco · 2026-11-18
- 27B Model Rewrites Its Own Weights Without Training or Gradients, Paper Claims — udmrzn · 2026-11-18
- From 'impossible' to a working Bluetooth beacon: open-source EDA chip design is real — IgorCarron · 2026-11-18
- AURA 2.0: a streaming video LLM that speaks first, calls tools, and grounds visually — Shuhan Li · 2026-11-18